Provided is an anomaly detection apparatus 10 capable of detecting anomalies of processing results. The anomaly detection apparatus 10 estimates first processing result information in accordance with first processing, using data regarding a target, compares second processing result data, for the data, obtained by second processing that differs from the first processing with the first processing result information, and detects an anomaly of the second processing according to the comparison result.
Legal claims defining the scope of protection, as filed with the USPTO.
1. An apparatus comprising: at least one memory storing instructions; and at least one processor configured to execute the instructions to: estimate high-level information of an upper layer of a DIKW (Data, Information, Knowledge and Wisdom) model from data of a lowest layer of the DIKW model, using first input data, wherein the first input data comprises observation data acquired from a camera or a lidar mounted on a robot; the DIKW model comprises a D (Data) layer as the lowest layer, an I (Information) layer above the D layer, a K (Knowledge) layer above the I layer, and a W (Wisdom) layer that is a highest layer and is above the K layer, the upper layer is the I, K, or W layer, the estimated high-level information corresponds to a user assessment of the first input data; input the first input data an artificial intelligence (AI) engine that executes predetermined processing; receive, as a processing result of the predetermined processing, output from the AI engine, the processing result corresponding to an AI assessment of the first input data; evaluate the processing result by comparing the processing result to the estimated high-level information; determine whether or not the processing result is abnormal according to an evaluation result, wherein the processing result is abnormal when the processing result differs from the estimated high-level information; adjust either or both of a control plan and a trajectory of the robot when the processing result has been determined as abnormal to reduce a likelihood that a future processing result based on future first input data is abnormal; and control the robot based on either or both of the control plan and the trajectory of the robot as has been adjusted.
2. The apparatus according to claim 1, wherein the at least one processor is configured to execute the instructions to further: estimate the high-level information by further using second input data that is different from the first input data and that is not input to the AI engine.
3. The apparatus according to claim 2, wherein the at least one processor is configured to execute the instructions to further: when the processing result is abnormal, output adjustment information used for adjusting the processing result of the AI engine.
4. The apparatus according to claim 3, wherein when a system has a plurality of AI engines, the evaluation, the determination and the adjustment processing are performed for each AI engine, and in the determination, the determination result for the processing result of the AI engine of a previous stage is input to the AI engine of a subsequent stage.
5. The apparatus according to claim 3, wherein when a system has a plurality of AI engines, the evaluation, the determination and the adjustment processing are performed for each AI engine, and in the evaluation, the high-level information that is common to each AI engine is used.
6. The apparatus according to claim 2, wherein the high-level information has a dimension lower than a dimension of the first and second input data, and is not a subset of the first and second input data.
7. The apparatus according to claim 1, wherein the high-level information is information representing a dependency between states of an environment by using at least a logical expression, a graph in graph theory, and a network.
8. The apparatus according to claim 1, wherein the at least one processor is configured to execute the instructions to further: compute a degree of conflict between the processing result and the high-level information, and wherein the processing result is a posterior probability distribution for the first input data, the high-level information is a probability distribution of a feature amount based on a precondition and a regularity, and the degree of conflict is evaluated by using any one of a likelihood, a KL (Kullback-Leibler) information amount, and distance information.
9. A method performed by an information processing device and comprising: estimating high-level information of an upper layer of a DIKW (Data, Information, Knowledge and Wisdom) model from data of a lowest layer of the DIKW model, using first input data, wherein the first input data comprises observation data acquired from a camera or a lidar mounted on a robot; the DIKW model comprises a D (Data) layer as the lowest layer, an I (Information) layer above the D layer, a K (Knowledge) layer above the I layer, and a W (Wisdom) layer that is a highest layer and is above the K layer, the upper layer is the I, K, or W layer, the estimated high-level information corresponds to a user assessment of the first input data; inputting the first input data an artificial intelligence (AI) engine that executes predetermined processing; receiving, as a processing result of the predetermined processing, output from the AI engine, the processing result corresponding to an AI assessment of the first input data; evaluating the processing result by comparing the processing result to the estimated high-level information; determining whether or not the processing result is abnormal according to an evaluation result, wherein the processing result is abnormal when the processing result differs from the estimated high-level information; adjusting either or both of a control plan and a trajectory of the robot when the processing result has been determined as abnormal to reduce a likelihood that a future processing result based on future first input data is abnormal; and controlling the robot based on either or both of the control plan and the trajectory of the robot as has been adjusted.
10. A non-transitory computer-readable recording medium storing a program executable by a computer to perform processing comprising: estimating high-level information of an upper layer of a DIKW (Data, Information, Knowledge and Wisdom) model from data of a lowest layer of the DIKW model, using first input data, wherein the first input data comprises observation data acquired from a camera or a lidar mounted on a robot; the DIKW model comprises a D (Data) layer as the lowest layer, an I (Information) layer above the D layer, a K (Knowledge) layer above the I layer, and a W (Wisdom) layer that is a highest layer and is above the K layer, the upper layer is the I, K, or W layer, the estimated high-level information corresponds to a user assessment of the first input data; inputting the first input data an artificial intelligence (AI) engine that executes predetermined processing; receiving, as a processing result of the predetermined processing, output from the AI engine, the processing result corresponding to an AI assessment of the first input data; evaluating the processing result by comparing the processing result to the estimated high-level information; determining whether or not the processing result is abnormal according to an evaluation result, wherein the processing result is abnormal when the processing result differs from the estimated high-level information; adjusting either or both of a control plan and a trajectory of the robot when the processing result has been determined as abnormal to reduce a likelihood that a future processing result based on future first input data is abnormal; and controlling the robot based on either or both of the control plan and the trajectory of the robot as has been adjusted.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
September 30, 2019
January 14, 2025
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.